eQTL follow up for key myositis driver SNPs (IMD basis PC1, 2, 3, 8, 9, 12, and 13)
Guillermo Reales
2021-12-06
Background
We want to follow-up a number of driver SNPs for our Myositis-relevant components in the IMD basis (PC1-3, 8, 9, 12, and 13). To do so, we retrieved eQTL summary statistics data from eQTL catalogue for the driver SNPs for these components. We further filtered those SNPs to include only those included in credible sets performed by eQTL catalogue using SuSIE, to be more sure that the SNPs are potentially causal. See below for more details on implementation.
First we load libraries and the list of SNPs
library(data.table)
library(readr)
library(seqminer)
library(magrittr)
library(ggplot2)
library(ggrepel)
library(cowplot)
library(gprofiler2)
# The following packages are Bioconductor's so they need to be installed differently
library(GenomicRanges)
library(biomaRt)
library(STRINGdb)
library(VennDiagram)
library(RColorBrewer)Load basic data and tissue selection
Here we’ll use data from the eQTLcatalogue, which we’ll additionally filter by information present in the credible sets computed by eQTL catalogue. This will help us tell which SNPs are truly causal eQTL for the genes and hence which eSNP-eGene are real (Note: Since latest update [June 2021] GTEx actually corresponds to GTEx v8, so GTEx_v8 is now redundant). Since I didn’t have a manifest for the credible paths (only URLs) I had to make some modifications in some names for both datasets to fit nicely. This involved some manual editing from the credible sets URLs, which resulted in credible_paths_processed.tsv.
We performed these steps for cell basis v3 (varimax) before, so some of the files are available in the corresponding directory.
# Let's import the relevant files, for report rendering.
snp <- fread("../data/Myositis_IMD_driver_SNPs.tsv")
setnames(snp, "SNPID", "SNPID.basis")
ecat <- fread("../data/Combined_eQTL_credible_paths.tsv") # This dataset was copied from blood cell basis v3 repository, since the sources are the sameNow let’s take a look at the available tissues
table(ecat$tissue_label)##
## adipose adipose (visceral)
## 12 4
## adrenal gland artery (aorta)
## 4 4
## artery (coronary) artery (tibial)
## 4 4
## B cell blood
## 6 12
## brain (amygdala) brain (anterior cingulate cortex)
## 4 4
## brain (caudate) brain (cerebellar hemisphere)
## 4 4
## brain (cerebellum) brain (cortex)
## 4 4
## brain (DLPFC) brain (hippocampus)
## 16 4
## brain (hypothalamus) brain (nucleus accumbens)
## 4 4
## brain (putamen) brain (spinal cord)
## 8 4
## brain (substantia nigra) brain (substantia_nigra)
## 4 4
## breast CD16+ monocyte
## 4 4
## CD4+ T cell CD8+ T cell
## 14 10
## esophagus (gej) esophagus (mucosa)
## 4 4
## esophagus (muscularis) fibroblast
## 4 8
## heart (atrial appendage) heart (left ventricle)
## 4 4
## hepatocyte high grade cartilage
## 4 4
## ileum iPSC
## 1 12
## kidney (cortex) LCL
## 4 24
## liver low grade cartilage
## 4 4
## lung macrophage
## 4 28
## microglia minor salivary gland
## 4 4
## monocyte muscle
## 33 8
## neutrophil NK cell
## 6 4
## ovary pancreas
## 4 4
## pancreatic islet pituitary
## 4 4
## placenta platelet
## 4 1
## prostate rectum
## 4 1
## sensory neuron sigmoid colon
## 4 4
## skin skin (suprapubic)
## 8 4
## small intestine spleen
## 4 4
## stomach synovium
## 4 4
## T cell testis
## 4 4
## Tfh cell Th1 cell
## 4 4
## Th1-17 cell Th17 cell
## 4 4
## Th2 cell thyroid
## 4 4
## tibial nerve transverse colon
## 4 5
## Treg memory Treg naive
## 4 4
## uterus vagina
## 4 4There are some interesting tissues to explore, particularly blood and immune cells, so we’ll select them. We’ll also select gene-expression and microarray datasets (since there are some interesting tissues, such as platelets, for which only microarray data is available), and “naive” condition.
tissues <- c("B cell", "CD16+ monocyte", "CD4+ T cell", "CD8+ T cell", "macrophage", "monocyte", "neutrophil", "NK cell", "platelet", "T cell", "Tfh cell", "Th1 cell", "Th1-17 cell", "Th17 cell", "Th2 cell", "Treg memory", "Treg naive")
ecat.filt <- ecat[tissue_label %in% tissues & (quant_method == "ge" | quant_method == "microarray") & condition_label == "naive",]At this point, we need to load the data. These files are reasonably big and their processing would take a while, so I decided to
- Download them all to the HPC (
~/rds/rds-cew54-wallace-share/Data/expr/eqtl-catalogue), - Filter them by our driver SNPs,
- Filter SNPs by those present in credible sets only,
- Retrieve extra information on associated genes from BiomaRt.
See eqtl_driverSNPs_processing.R to get a taste of what I did.
Let’s load the full resulting dataset.
causal <- fread("../data/Causal_eQTLs_full_20210915.tsv") # File copied from IMD basis repository, as it contains the required information
# Noticed a bug in the procesing of snp_study and study_tissue. Let's briefly fix it
causal[, snp_study:=gsub(pattern = "(_[A-Za-z0-9\\+ -]+)$", "", snp_study, perl = TRUE)][, study_tissue:=gsub(pattern = "(_[A-Za-z0-9\\+ -]+)$", "", study_tissue, perl = TRUE)][,study:=gsub(pattern = "(_[A-Za-z0-9\\+ -]+)$", "", study, perl = TRUE)][,quant_method:=gsub("([A-Za-z0-9_]+)_([a-z]+)", "\\2", study, perl = TRUE)][,study:=gsub(pattern = "(_[a-z]+)$", "", study, perl = TRUE)]Data vizzzzz
Some EDA – Which genes and tissues are more popular?
We have multiple genes across different tissues, let’s check which tissues contain more eQTL-Gene associations.
causal[, uniqueN(gene_name), by=tissue_label][order(-V1)]Monocytes seem to have the most eQTL-gene associations. But is it also considered in more studies than other tissues?
table(causal$study, causal$tissue_label)##
## B cell CD16+ monocyte CD4+ T cell CD8+ T cell macrophage
## Alasoo_2018 0 0 0 0 10
## BLUEPRINT 0 0 26 0 0
## CEDAR 5 0 11 6 0
## Fairfax_2012 26 0 0 0 0
## Fairfax_2014 0 0 0 0 0
## GENCORD 0 0 0 0 0
## Kasela_2017 0 0 11 8 0
## Naranbhai_2015 0 0 0 0 0
## Nedelec_2016 0 0 0 0 4
## Quach_2016 0 0 0 0 0
## Schmiedel_2018 11 9 9 14 0
##
## monocyte neutrophil NK cell platelet T cell Tfh cell Th1 cell
## Alasoo_2018 0 0 0 0 0 0 0
## BLUEPRINT 45 35 0 0 0 0 0
## CEDAR 3 7 0 4 0 0 0
## Fairfax_2012 0 0 0 0 0 0 0
## Fairfax_2014 13 0 0 0 0 0 0
## GENCORD 0 0 0 0 16 0 0
## Kasela_2017 0 0 0 0 0 0 0
## Naranbhai_2015 0 10 0 0 0 0 0
## Nedelec_2016 0 0 0 0 0 0 0
## Quach_2016 5 0 0 0 0 0 0
## Schmiedel_2018 16 0 6 0 0 7 15
##
## Th1-17 cell Th17 cell Th2 cell Treg memory Treg naive
## Alasoo_2018 0 0 0 0 0
## BLUEPRINT 0 0 0 0 0
## CEDAR 0 0 0 0 0
## Fairfax_2012 0 0 0 0 0
## Fairfax_2014 0 0 0 0 0
## GENCORD 0 0 0 0 0
## Kasela_2017 0 0 0 0 0
## Naranbhai_2015 0 0 0 0 0
## Nedelec_2016 0 0 0 0 0
## Quach_2016 0 0 0 0 0
## Schmiedel_2018 12 14 14 14 10So it seems!
summ <- causal[,.(genes = uniqueN(gene_name), studies = uniqueN(study)), by=tissue_label][order(-genes)]
plot(summ$studies, summ$genes, xlab = "Number of studies", ylab="Number of unique genes")
So there seems to be a relationship between how many studies the tissue was studied in and the number of eQTL-gene associations, but our sample is small and very skewed, as most tissues were studied in one study only (ie. Schmiedel et al., 2018 included several cell types, but was alone in our catalogue in doing so).
Which genes are more popular across tissues?
summ2 <- causal[, uniqueN(tissue_label), by=gene_name][order(-V1)]
summ2hist(summ2$V1, breaks = 1:17)
Overall most eQTL-gene associations seem to be restricted to one or few tissues.
No gene seems to be represented across all tissues, but some are represented in 11/17 tissues which are ORMDL3, which has been associated with asthma, and RAB5C, associated with autoimmune diseases (ie vitiligo) and monocyte counts.
Principal components
We’ll focus on the different driver SNPs component by component.
PC1
Driver SNPs and cell types
We’ll take a look at how different cell types are influenced by driver SNPs at PC1.
snppc1 <- snp[rot1 != 0]
pc1 <- causal[rot1 != 0]
pc1[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot1)]
snppc1[, eqtl:=ifelse(SNPID.basis %in% pc1$SNPID.basis, "yes", "no")]
pc1dp <- ggplot(snppc1, aes(x = reorder(SNPID.basis, rot1), y = rot1, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc1[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC1 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc1dp
length(unique(pc1$SNPID.basis))## [1] 15length(unique(pc1$tissue_label))## [1] 11For PC1 we have 15 driver SNPs that are identified as eQTL across 11 tissues.
tcells <- c("CD4+ T cell", "CD8+ T cell", "T cell", "Tfh cell", "Th1 cell", "Th1-17 cell", "Th17 cell", "Th2 cell", "Treg memory", "Treg naive")
otcells <- c("B cell", "CD16+ monocyte", "macrophage", "monocyte", "neutrophil", "NK cell", "platelet")
plot.cells.eqtl <- function(x){
tcells <- c("CD4+ T cell", "CD8+ T cell", "T cell", "Tfh cell", "Th1 cell", "Th1-17 cell", "Th17 cell", "Th2 cell", "Treg memory", "Treg naive")
otcells <- c("B cell", "CD16+ monocyte", "macrophage", "monocyte", "neutrophil", "NK cell", "platelet")
ptc <- x[tissue_label %in% tcells]
poc <- x[tissue_label %in% otcells]
cp1 <- ggplot(ptc, aes(x = SNPID.basis, y = beta, ymin=beta-se, ymax=beta+se, colour = gene_name))+
geom_pointrange()+
geom_hline(yintercept = 0, col="red", lty=2)+
ylab("Beta")+
scale_x_discrete(drop=FALSE)+
ylim(c(min(x$beta), max(x$beta)))+
# ggtitle("PC1 driver SNPs effect on gene expression")+
facet_grid(tissue_label~., scales = "free", space = "free", switch = "y")+
theme_bw(base_size = 18)+
theme(legend.position = "none", strip.text.y.left = element_text(angle = 0), axis.title.y = element_blank(), axis.text.x = element_text(angle = 90))
cp2 <- ggplot(poc, aes(x = SNPID.basis, y = beta, ymin=beta-se, ymax=beta+se, colour = gene_name))+
geom_pointrange()+
geom_hline(yintercept = 0, col="red", lty=2)+
ylab("Beta")+
scale_x_discrete(drop=FALSE)+
ylim(c(min(x$beta), max(x$beta)))+
# ggtitle("PC1 driver SNPs effect on gene expression")+
facet_grid(tissue_label~., scales = "free", space = "free", switch = "y")+
theme_bw(base_size = 18)+
theme(legend.position = "bottom", strip.text.y.left = element_text(angle = 0), axis.title.y = element_blank(), axis.text.x = element_text(angle = 90))
gridExtra::grid.arrange(cp1, cp2, nrow = 1)
}
pce1 <- plot.cells.eqtl(pc1)## Warning: Removed 6 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
PC2
Driver SNPs and cell types
snppc2 <- snp[rot2 != 0]
pc2 <- causal[rot2 != 0]
pc2[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot2)]
snppc2[, eqtl:=ifelse(SNPID.basis %in% pc2$SNPID.basis, "yes", "no")]
pc2dp <- ggplot(snppc2, aes(x = reorder(SNPID.basis, rot2), y = rot2, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc2[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC2 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc2dp
length(unique(pc2$SNPID.basis))## [1] 24length(unique(pc2$tissue_label))## [1] 17plot.cells.eqtl(pc2)## Warning: Removed 7 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
PC3
Driver SNPs and cell types
snppc3 <- snp[rot3 != 0]
pc3 <- causal[rot3 != 0]
pc3[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot3)]
snppc3[, eqtl:=ifelse(SNPID.basis %in% pc3$SNPID.basis, "yes", "no")]
pc3dp <- ggplot(snppc3, aes(x = reorder(SNPID.basis, rot3), y = rot3, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc3[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC3 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc3dp
length(unique(pc3$SNPID.basis))## [1] 35length(unique(pc3$tissue_label))## [1] 17plot.cells.eqtl(pc3)## Warning: Removed 9 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).## Warning: Removed 2 rows containing missing values (geom_segment).
PC8
Driver SNPs and cell types
snppc8 <- snp[rot8 != 0]
pc8 <- causal[rot8 != 0]
pc8[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot8)]
snppc8[, eqtl:=ifelse(SNPID.basis %in% pc8$SNPID.basis, "yes", "no")]
pc8dp <- ggplot(snppc8, aes(x = reorder(SNPID.basis, rot8), y = rot8, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc8[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC8 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc8dp
length(unique(pc8$SNPID.basis))## [1] 26length(unique(pc8$tissue_label))## [1] 14plot.cells.eqtl(pc8)## Warning: Removed 5 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
PC9
Driver SNPs and cell types
snppc9 <- snp[rot9 != 0]
pc9 <- causal[rot9 != 0]
pc9[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot9)]
snppc9[, eqtl:=ifelse(SNPID.basis %in% pc9$SNPID.basis, "yes", "no")]
pc9dp <- ggplot(snppc9, aes(x = reorder(SNPID.basis, rot9), y = rot9, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc9[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC9 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc9dp
length(unique(pc9$SNPID.basis))## [1] 54length(unique(pc9$tissue_label))## [1] 17plot.cells.eqtl(pc9)## Warning: Removed 17 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
PC12
Driver SNPs and cell types
snppc12 <- snp[rot12 != 0]
pc12 <- causal[rot12 != 0]
pc12[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot12)]
snppc12[, eqtl:=ifelse(SNPID.basis %in% pc12$SNPID.basis, "yes", "no")]
pc12dp <- ggplot(snppc12, aes(x = reorder(SNPID.basis, rot12), y = rot12, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc12[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC12 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc12dp
length(unique(pc12$SNPID.basis))## [1] 51length(unique(pc12$tissue_label))## [1] 17plot.cells.eqtl(pc12)## Warning: Removed 16 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
PC13
Driver SNPs and cell types
snppc13 <- snp[rot12 != 0]
pc13 <- causal[rot12 != 0]
pc13[, SNPID.basis:=factor(SNPID.basis)][, SNPID.basis:= reorder(SNPID.basis, rot13)]
snppc13[, eqtl:=ifelse(SNPID.basis %in% pc13$SNPID.basis, "yes", "no")]
pc13dp <- ggplot(snppc13, aes(x = reorder(SNPID.basis, rot13), y = rot13, colour = eqtl, label=SNPID.basis))+
geom_point()+
scale_colour_manual(values = c("yes" = "red", "no" = "black"))+
geom_hline(yintercept = 0, col="red", lty=1)+
geom_label_repel(size = 3, force = 2, box.padding = .3, seed = 2, direction = "both", max.overlaps = 45, data = snppc13[eqtl == "yes"])+
ylab("Driver SNP Rotation")+
ggtitle("PC13 driver SNPs")+
theme_cowplot(12)+
theme(axis.text.x = element_text(angle = 90, hjust = 1), axis.title.x = element_blank(), axis.text.x.bottom = element_blank())
pc13dp
length(unique(pc13$SNPID.basis))## [1] 51length(unique(pc13$tissue_label))## [1] 17plot.cells.eqtl(pc13)## Warning: Removed 16 rows containing missing values (geom_segment).## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
## Warning: Removed 1 rows containing missing values (geom_segment).
–>
–> –> –> –> –>
–>
–>
–>
–> –> –> –>
–> –> –> –> –>
–>
–>
–> –>
–>
–> –> –>
–>
–>
–>
–>
–> –> –> –> –> –>
–>
–> –> –> –>
–>
–> –> –> –>
–>
–> –> –> –>
–>
–> –> –> –>